2007-11-04 23:24:55

by Mathieu Desnoyers

[permalink] [raw]
Subject: [to-be-posted-soon] Multiple handlers per marker

* Mathieu Desnoyers ([email protected]) wrote:
> * Mathieu Desnoyers ([email protected]) wrote:
> > * Mike Mason ([email protected]) wrote:
> > > Hi Mathieu,
> > >
> > > Are you aware of any working being done to allow multiple handlers to be
> > > attached to a marker? Something like what kprobes allows. I've started to
> > > look into this and don't want to duplicate efforts.
> > >
> >
> > Nope, but I know we will have to address this.
> >
> > Something along the lines of walking an RCU list of function pointers,
> > calling them.
> >
> > The only downside I see is that we will have to pass a va_list * instead
> > of real va args. The could make the marker site a little bit bigger and
> > will change the probe callback arguments.
> >
> > What do you think about these ideas ?
> >
> > If we can find a way to make the common case (only one probe connected)
> > _ultra_ fast, and yet architecture independent, that would be awesome. A
> > simple call is kind of hard to beat though.. So we may have to think
> > about a design with :
> >
> > - One call at the marker site
> > - if 1 probe is installed :
> > - If the format string is empty, connect a probe without va args.
> > - If the format string is not empty, connect a "stage 1" probe that takes
> > the va args, starts/ends the va_list and calls _one_ function (let's
> > call it "stage 2" probe), that takes va_list as parameter.
> > - if more than 1 probe is installed :
> > - The stage 1 probe creates the va_list and passes it to each function
> > connected, iterated with an RCU list.
> >
> > What do you think ?
> >
> > Mathieu
> >
>
> I'm working on an implementation.
>

It's ready for testing. Please grab
http://ltt.polymtl.ca/lttng/patch-2.6.24-rc1-git13-lttng-0.10-pre18.tar.bz2
patch name :

markers-support-multiple-probes.patch

It still need to go through patchcheck.pl and some polishing, but it
seems to work fine for me with multiple probes (the sample marker,
sample probe and multiple instances of my lttng probes can
connect/disconnect without problem).

Currently, the "connect/disconnect" and "arm/disarm" operations are
separate. However, they could be merged. Any comment/preference on this?
Being separate, a probe provider can wait until the very last moment
before it activates its markers, with a minimalistic impact on the
system, but it is not such a strong argument.

Mathieu

--
Mathieu Desnoyers
Computer Engineering Ph.D. Student, Ecole Polytechnique de Montreal
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F BA06 3F25 A8FE 3BAE 9A68


2007-11-05 23:14:33

by Mike Mason

[permalink] [raw]
Subject: Re: [to-be-posted-soon] Multiple handlers per marker

Mathieu Desnoyers wrote:
> * Mathieu Desnoyers ([email protected]) wrote:
>> * Mathieu Desnoyers ([email protected]) wrote:
>>> * Mike Mason ([email protected]) wrote:
>>>> Hi Mathieu,
>>>>
>>>> Are you aware of any working being done to allow multiple handlers to be
>>>> attached to a marker? Something like what kprobes allows. I've started to
>>>> look into this and don't want to duplicate efforts.
>>>>
>>> Nope, but I know we will have to address this.
>>>
>>> Something along the lines of walking an RCU list of function pointers,
>>> calling them.
>>>
>>> The only downside I see is that we will have to pass a va_list * instead
>>> of real va args. The could make the marker site a little bit bigger and
>>> will change the probe callback arguments.
>>>
>>> What do you think about these ideas ?
>>>
>>> If we can find a way to make the common case (only one probe connected)
>>> _ultra_ fast, and yet architecture independent, that would be awesome. A
>>> simple call is kind of hard to beat though.. So we may have to think
>>> about a design with :
>>>
>>> - One call at the marker site
>>> - if 1 probe is installed :
>>> - If the format string is empty, connect a probe without va args.
>>> - If the format string is not empty, connect a "stage 1" probe that takes
>>> the va args, starts/ends the va_list and calls _one_ function (let's
>>> call it "stage 2" probe), that takes va_list as parameter.
>>> - if more than 1 probe is installed :
>>> - The stage 1 probe creates the va_list and passes it to each function
>>> connected, iterated with an RCU list.
>>>
>>> What do you think ?
>>>
>>> Mathieu
>>>
>> I'm working on an implementation.
>>
>
> It's ready for testing. Please grab
> http://ltt.polymtl.ca/lttng/patch-2.6.24-rc1-git13-lttng-0.10-pre18.tar.bz2
> patch name :
>
> markers-support-multiple-probes.patch

This patch alone doesn't apply cleanly at all on 2.6.24-rc1-git14. Are there other patches in this series I should apply first?

Mike

>
> It still need to go through patchcheck.pl and some polishing, but it
> seems to work fine for me with multiple probes (the sample marker,
> sample probe and multiple instances of my lttng probes can
> connect/disconnect without problem).
>
> Currently, the "connect/disconnect" and "arm/disarm" operations are
> separate. However, they could be merged. Any comment/preference on this?
> Being separate, a probe provider can wait until the very last moment
> before it activates its markers, with a minimalistic impact on the
> system, but it is not such a strong argument.
>
> Mathieu
>

2007-11-05 23:17:52

by Mathieu Desnoyers

[permalink] [raw]
Subject: Re: [to-be-posted-soon] Multiple handlers per marker

* Mike Mason ([email protected]) wrote:
> Mathieu Desnoyers wrote:
>> * Mathieu Desnoyers ([email protected]) wrote:
>>> * Mathieu Desnoyers ([email protected]) wrote:
>>>> * Mike Mason ([email protected]) wrote:
>>>>> Hi Mathieu,
>>>>>
>>>>> Are you aware of any working being done to allow multiple handlers to
>>>>> be attached to a marker? Something like what kprobes allows. I've
>>>>> started to look into this and don't want to duplicate efforts.
>>>>>
>>>> Nope, but I know we will have to address this.
>>>>
>>>> Something along the lines of walking an RCU list of function pointers,
>>>> calling them.
>>>>
>>>> The only downside I see is that we will have to pass a va_list * instead
>>>> of real va args. The could make the marker site a little bit bigger and
>>>> will change the probe callback arguments.
>>>>
>>>> What do you think about these ideas ?
>>>>
>>>> If we can find a way to make the common case (only one probe connected)
>>>> _ultra_ fast, and yet architecture independent, that would be awesome. A
>>>> simple call is kind of hard to beat though.. So we may have to think
>>>> about a design with :
>>>>
>>>> - One call at the marker site
>>>> - if 1 probe is installed :
>>>> - If the format string is empty, connect a probe without va args.
>>>> - If the format string is not empty, connect a "stage 1" probe that
>>>> takes
>>>> the va args, starts/ends the va_list and calls _one_ function (let's
>>>> call it "stage 2" probe), that takes va_list as parameter.
>>>> - if more than 1 probe is installed :
>>>> - The stage 1 probe creates the va_list and passes it to each function
>>>> connected, iterated with an RCU list.
>>>>
>>>> What do you think ?
>>>>
>>>> Mathieu
>>>>
>>> I'm working on an implementation.
>>>
>> It's ready for testing. Please grab
>> http://ltt.polymtl.ca/lttng/patch-2.6.24-rc1-git13-lttng-0.10-pre18.tar.bz2
>> patch name :
>> markers-support-multiple-probes.patch
>
> This patch alone doesn't apply cleanly at all on 2.6.24-rc1-git14. Are
> there other patches in this series I should apply first?
>

Yes, the following ones should suffice :

# instrumentation menu removal
add-kconfig-to-arch.patch
add-arch-supports-oprofile.patch
add-arch-supports-kprobes.patch
move-kconfig-instrumentation-to-arch.patch
#
kprobes-use-mutex-for-insn-pages.patch
kprobes-dont-use-kprobes-mutex-in-arch-code.patch
kprobes-declare-kprobes-mutex-static.patch
declare-array.patch
text-edit-lock-architecture-independent-code.patch
text-edit-lock-alternative-i386-and-x86_64.patch
text-edit-lock-kprobes-architecture-independent.patch
text-edit-lock-kprobes-i386.patch
text-edit-lock-kprobes-x86_64.patch
text-edit-lock-i386-standardize-debug-rodata.patch
text-edit-lock-x86_64-standardize-debug-rodata.patch
#
immediate-values-architecture-independent-code.patch
immediate-values-kconfig-embedded.patch
immediate-values-move-kprobes-i386-restore-interrupt-to-kdebug-h.patch
add-asm-compat-to-x86.patch
immediate-values-i386-optimization.patch
immediate-values-powerpc-optimization.patch
immediate-values-documentation.patch
#
linux-kernel-markers-immediate-values.patch
#
markers-support-multiple-probes.patch

Tell me if you still have rejects.

Mathieu


> Mike
>
>> It still need to go through patchcheck.pl and some polishing, but it
>> seems to work fine for me with multiple probes (the sample marker,
>> sample probe and multiple instances of my lttng probes can
>> connect/disconnect without problem).
>> Currently, the "connect/disconnect" and "arm/disarm" operations are
>> separate. However, they could be merged. Any comment/preference on this?
>> Being separate, a probe provider can wait until the very last moment
>> before it activates its markers, with a minimalistic impact on the
>> system, but it is not such a strong argument.
>> Mathieu
>

--
Mathieu Desnoyers
Computer Engineering Ph.D. Student, Ecole Polytechnique de Montreal
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F BA06 3F25 A8FE 3BAE 9A68

2007-11-06 22:37:24

by Mike Mason

[permalink] [raw]
Subject: Re: [to-be-posted-soon] Multiple handlers per marker

Mathieu Desnoyers wrote:
> * Mike Mason ([email protected]) wrote:
>> Mathieu Desnoyers wrote:
>>> * Mathieu Desnoyers ([email protected]) wrote:
>>>> * Mathieu Desnoyers ([email protected]) wrote:
>>>>> * Mike Mason ([email protected]) wrote:
>>>>>> Hi Mathieu,
>>>>>>
>>>>>> Are you aware of any working being done to allow multiple handlers to
>>>>>> be attached to a marker? Something like what kprobes allows. I've
>>>>>> started to look into this and don't want to duplicate efforts.
>>>>>>
>>>>> Nope, but I know we will have to address this.
>>>>>
>>>>> Something along the lines of walking an RCU list of function pointers,
>>>>> calling them.
>>>>>
>>>>> The only downside I see is that we will have to pass a va_list * instead
>>>>> of real va args. The could make the marker site a little bit bigger and
>>>>> will change the probe callback arguments.
>>>>>
>>>>> What do you think about these ideas ?
>>>>>
>>>>> If we can find a way to make the common case (only one probe connected)
>>>>> _ultra_ fast, and yet architecture independent, that would be awesome. A
>>>>> simple call is kind of hard to beat though.. So we may have to think
>>>>> about a design with :
>>>>>
>>>>> - One call at the marker site
>>>>> - if 1 probe is installed :
>>>>> - If the format string is empty, connect a probe without va args.
>>>>> - If the format string is not empty, connect a "stage 1" probe that
>>>>> takes
>>>>> the va args, starts/ends the va_list and calls _one_ function (let's
>>>>> call it "stage 2" probe), that takes va_list as parameter.
>>>>> - if more than 1 probe is installed :
>>>>> - The stage 1 probe creates the va_list and passes it to each function
>>>>> connected, iterated with an RCU list.
>>>>>
>>>>> What do you think ?
>>>>>
>>>>> Mathieu
>>>>>
>>>> I'm working on an implementation.
>>>>
>>> It's ready for testing. Please grab
>>> http://ltt.polymtl.ca/lttng/patch-2.6.24-rc1-git13-lttng-0.10-pre18.tar.bz2
>>> patch name :
>>> markers-support-multiple-probes.patch
>> This patch alone doesn't apply cleanly at all on 2.6.24-rc1-git14. Are
>> there other patches in this series I should apply first?
>>
>
> Yes, the following ones should suffice :
>
> # instrumentation menu removal
> add-kconfig-to-arch.patch
> add-arch-supports-oprofile.patch
> add-arch-supports-kprobes.patch
> move-kconfig-instrumentation-to-arch.patch
> #
> kprobes-use-mutex-for-insn-pages.patch
> kprobes-dont-use-kprobes-mutex-in-arch-code.patch
> kprobes-declare-kprobes-mutex-static.patch
> declare-array.patch
> text-edit-lock-architecture-independent-code.patch
> text-edit-lock-alternative-i386-and-x86_64.patch
> text-edit-lock-kprobes-architecture-independent.patch
> text-edit-lock-kprobes-i386.patch
> text-edit-lock-kprobes-x86_64.patch
> text-edit-lock-i386-standardize-debug-rodata.patch
> text-edit-lock-x86_64-standardize-debug-rodata.patch
> #
> immediate-values-architecture-independent-code.patch
> immediate-values-kconfig-embedded.patch
> immediate-values-move-kprobes-i386-restore-interrupt-to-kdebug-h.patch
> add-asm-compat-to-x86.patch
> immediate-values-i386-optimization.patch
> immediate-values-powerpc-optimization.patch
> immediate-values-documentation.patch
> #
> linux-kernel-markers-immediate-values.patch
> #
> markers-support-multiple-probes.patch
>
> Tell me if you still have rejects.

I applied the above patches to 2.6.24-rc1-git14. They applied fine with just a few offsets until the last patch, which yielded this result:

patching file include/linux/marker.h
Hunk #5 succeeded at 162 with fuzz 2.
patching file kernel/marker.c
Hunk #14 FAILED at 534.
Hunk #15 FAILED at 587.
Hunk #16 FAILED at 621.
Hunk #17 FAILED at 732.
Hunk #18 FAILED at 769.
Hunk #19 succeeded at 791 (offset 12 lines).
5 out of 19 hunks FAILED -- saving rejects to file kernel/marker.c.rej
patching file kernel/module.c
Hunk #1 succeeded at 1998 (offset -3 lines).
Hunk #2 succeeded at 2608 (offset -37 lines).
Hunk #3 succeeded at 2651 with fuzz 1 (offset -3 lines).
patching file include/linux/module.h
Hunk #1 FAILED at 468.
Hunk #2 succeeded at 572 (offset -2 lines).
1 out of 2 hunks FAILED -- saving rejects to file include/linux/module.h.rej
patching file samples/markers/probe-example.c

Mike

>
> Mathieu
>
>
>> Mike
>>
>>> It still need to go through patchcheck.pl and some polishing, but it
>>> seems to work fine for me with multiple probes (the sample marker,
>>> sample probe and multiple instances of my lttng probes can
>>> connect/disconnect without problem).
>>> Currently, the "connect/disconnect" and "arm/disarm" operations are
>>> separate. However, they could be merged. Any comment/preference on this?
>>> Being separate, a probe provider can wait until the very last moment
>>> before it activates its markers, with a minimalistic impact on the
>>> system, but it is not such a strong argument.
>>> Mathieu
>

2007-11-07 16:55:27

by Mathieu Desnoyers

[permalink] [raw]
Subject: Re: [to-be-posted-soon] Multiple handlers per marker

* Mike Mason ([email protected]) wrote:
> Mathieu Desnoyers wrote:
>> * Mike Mason ([email protected]) wrote:
>>> Mathieu Desnoyers wrote:
>>>> * Mathieu Desnoyers ([email protected]) wrote:
>>>>> * Mathieu Desnoyers ([email protected]) wrote:
>>>>>> * Mike Mason ([email protected]) wrote:
>>>>>>> Hi Mathieu,
>>>>>>>
>>>>>>> Are you aware of any working being done to allow multiple handlers to
>>>>>>> be attached to a marker? Something like what kprobes allows. I've
>>>>>>> started to look into this and don't want to duplicate efforts.
>>>>>>>
>>>>>> Nope, but I know we will have to address this.
>>>>>>
>>>>>> Something along the lines of walking an RCU list of function pointers,
>>>>>> calling them.
>>>>>>
>>>>>> The only downside I see is that we will have to pass a va_list *
>>>>>> instead
>>>>>> of real va args. The could make the marker site a little bit bigger
>>>>>> and
>>>>>> will change the probe callback arguments.
>>>>>>
>>>>>> What do you think about these ideas ?
>>>>>>
>>>>>> If we can find a way to make the common case (only one probe
>>>>>> connected)
>>>>>> _ultra_ fast, and yet architecture independent, that would be awesome.
>>>>>> A
>>>>>> simple call is kind of hard to beat though.. So we may have to think
>>>>>> about a design with :
>>>>>>
>>>>>> - One call at the marker site
>>>>>> - if 1 probe is installed :
>>>>>> - If the format string is empty, connect a probe without va args.
>>>>>> - If the format string is not empty, connect a "stage 1" probe that
>>>>>> takes
>>>>>> the va args, starts/ends the va_list and calls _one_ function
>>>>>> (let's
>>>>>> call it "stage 2" probe), that takes va_list as parameter.
>>>>>> - if more than 1 probe is installed :
>>>>>> - The stage 1 probe creates the va_list and passes it to each
>>>>>> function
>>>>>> connected, iterated with an RCU list.
>>>>>>
>>>>>> What do you think ?
>>>>>>
>>>>>> Mathieu
>>>>>>
>>>>> I'm working on an implementation.
>>>>>
>>>> It's ready for testing. Please grab
>>>> http://ltt.polymtl.ca/lttng/patch-2.6.24-rc1-git13-lttng-0.10-pre18.tar.bz2
>>>> patch name :
>>>> markers-support-multiple-probes.patch
>>> This patch alone doesn't apply cleanly at all on 2.6.24-rc1-git14. Are
>>> there other patches in this series I should apply first?
>>>
>> Yes, the following ones should suffice :
>> # instrumentation menu removal
>> add-kconfig-to-arch.patch
>> add-arch-supports-oprofile.patch
>> add-arch-supports-kprobes.patch
>> move-kconfig-instrumentation-to-arch.patch
>> #
>> kprobes-use-mutex-for-insn-pages.patch
>> kprobes-dont-use-kprobes-mutex-in-arch-code.patch
>> kprobes-declare-kprobes-mutex-static.patch
>> declare-array.patch
>> text-edit-lock-architecture-independent-code.patch
>> text-edit-lock-alternative-i386-and-x86_64.patch
>> text-edit-lock-kprobes-architecture-independent.patch
>> text-edit-lock-kprobes-i386.patch
>> text-edit-lock-kprobes-x86_64.patch
>> text-edit-lock-i386-standardize-debug-rodata.patch
>> text-edit-lock-x86_64-standardize-debug-rodata.patch
>> #
>> immediate-values-architecture-independent-code.patch
>> immediate-values-kconfig-embedded.patch
>> immediate-values-move-kprobes-i386-restore-interrupt-to-kdebug-h.patch
>> add-asm-compat-to-x86.patch
>> immediate-values-i386-optimization.patch
>> immediate-values-powerpc-optimization.patch
>> immediate-values-documentation.patch
>> #
>> linux-kernel-markers-immediate-values.patch
>> #
>> markers-support-multiple-probes.patch
>> Tell me if you still have rejects.
>
> I applied the above patches to 2.6.24-rc1-git14. They applied fine with
> just a few offsets until the last patch, which yielded this result:
>
> patching file include/linux/marker.h
> Hunk #5 succeeded at 162 with fuzz 2.
> patching file kernel/marker.c
> Hunk #14 FAILED at 534.
> Hunk #15 FAILED at 587.
> Hunk #16 FAILED at 621.
> Hunk #17 FAILED at 732.
> Hunk #18 FAILED at 769.
> Hunk #19 succeeded at 791 (offset 12 lines).
> 5 out of 19 hunks FAILED -- saving rejects to file kernel/marker.c.rej
> patching file kernel/module.c
> Hunk #1 succeeded at 1998 (offset -3 lines).
> Hunk #2 succeeded at 2608 (offset -37 lines).
> Hunk #3 succeeded at 2651 with fuzz 1 (offset -3 lines).
> patching file include/linux/module.h
> Hunk #1 FAILED at 468.
> Hunk #2 succeeded at 572 (offset -2 lines).
> 1 out of 2 hunks FAILED -- saving rejects to file
> include/linux/module.h.rej
> patching file samples/markers/probe-example.c
>
> Mike
>

Ok, I released a new patchset, which should fix your problem :

http://ltt.polymtl.ca/lttng/patch-2.6.24-rc2-lttng-0.10-pre20.tar.bz2

You simply have to apply all patches up to

markers-support-multiple-probes.patch

I have moved the patch earlier in the patchset so you don't have to
apply lttng. I also fixed the coding style and bugs I encountered during
my testing. You may also want to try out
markers-multi-probes-test.patch, which is a test module that I used to
make sure the probes were correct upon multiple connect/disconnect. It
is useful when you activate the "marker_debug" integer in
kernel/marker.c.

For those interested in lttng, this version should be used with :
http://ltt.polymtl.ca/lttng/ltt-control-0.46-06112007.tar.gz
http://ltt.polymtl.ca/packages/lttv-0.10.0-pre2-07112007.tar.gz

Mathieu

>> Mathieu
>>> Mike
>>>
>>>> It still need to go through patchcheck.pl and some polishing, but it
>>>> seems to work fine for me with multiple probes (the sample marker,
>>>> sample probe and multiple instances of my lttng probes can
>>>> connect/disconnect without problem).
>>>> Currently, the "connect/disconnect" and "arm/disarm" operations are
>>>> separate. However, they could be merged. Any comment/preference on this?
>>>> Being separate, a probe provider can wait until the very last moment
>>>> before it activates its markers, with a minimalistic impact on the
>>>> system, but it is not such a strong argument.
>>>> Mathieu
>

--
Mathieu Desnoyers
Computer Engineering Ph.D. Student, Ecole Polytechnique de Montreal
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F BA06 3F25 A8FE 3BAE 9A68